Partial Order Pruning: For Best Speed/Accuracy Trade-Off in Neural Architecture Search

By Xin Li, Yiming Zhou, Zheng Pan, Jiashi Feng. 2019

173 citations recorded by the source corpus.

Achieving good speed and accuracy trade-off on a target platform is very important in deploying deep neural networks in real world scenarios. However, most existing automatic architecture search approaches only concentrate on high performance. In this work, we propose an algorithm that can offer better speed/accuracy trade-off of searched networks, which is termed "Partial Order Pruning''. It prunes the architecture search space with a partial order assumption to automatically search for the architectures with the best speed and accuracy trade-off. Our algorithm explicitly takes profile information about the inference speed on the target platform into consideration. With the proposed algorithm, we present several Dongfeng (DF) networks that provide high accuracy and fast inference speed on various application GPU platforms. By further searching decoder architectures, our DF-Seg real-time segmentation networks yield state-of-the-art speed/accuracy trade-off on both the {target embedded device} and the high-end GPU.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xiaodan Liang
2. Alan L. Yuille
3. Yi Yang
4. Xiaojun Chang
5. Jingdong Wang
6. Xuanyi Dong
7. Lu Yuan
8. Zhangyang Wang

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Yi Yang
3. Dacheng Tao
4. Ming-Hsuan Yang
5. Xiaogang Wang
6. Alan L. Yuille
7. Chunhua Shen
8. Xiaodan Liang

Citations in the corpus

• A Comprehensive Survey of Neural Architecture Search (2020, background): “In addition, there is some work to consider how to strike a balance between performance and efficiency [116, 119].”
• Searching for a Robust Neural Architecture in Four GPU Hours (2019, background): “Macro search algorithms aim to directly discover the entire neural networks [5, 4, 38, 46, 21].”
• PIDNet: A Real-time Semantic Segmentation Network Inspired by PID Controllers (2022, unknown): “2 GTX 1080Ti 1536×768 DF2-Seg2 [29] 76.”
• CLIP-Driven Universal Model for Organ Segmentation and Tumor Detection (2023, background): “[37] Xin Li, Yiming Zhou, Zheng Pan, and Jiashi Feng.”
• UNet++: Redesigning Skip Connections to Exploit Multiscale Features in Image Segmentation (2019, unknown): “hindered by the limited computational resources [19]–[23].”
• Lite-HRNet: A Lightweight High-Resolution Network (2021, methodology): “02 GFLOPs, outperforming the hand-crafted methods [34, 55, 50, 23, 58, 36, 41, 12] and NAS-based methods [54, 25, 16, 24] , and comparable with SwiftNetRN18 [32] that is far computationally intensive (104 GFLOPs).”
• MTL-NAS: Task-Agnostic Neural Architecture Search Towards General-Purpose Multi-Task Learning (2020, background): “Recently, many neural architecture search (NAS) methods have emerged to jointly learn the weights and the network architectures [4, 6, 12, 14, 22, 31, 34, 41, 44, 45, 71].”
• SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (2020, background): “Firstly, the input image is feed into a stem architecture following [16, 42], which consists of two 3 × 3 convolutions (stride=2) decreasing the resolution to and the number of base channels in main body is equal to the number of output channels of the stem.”
• Multi-objective Neural Architecture Search via Predictive Network Performance Optimization (2019, unknown): “Params Top-1 err No. of samples truly evaluated NASNet-A+cutout (Zoph et al., 2018) 3.3 M 2.65 20000 AmoebaNet-B+cutout (Real et al., 2019b) 2.8 M 2.55 27000 PNASNet-5 (Liu et al., 2018a) 3.2 M 3.41 1160 NAO (Luo et al., 2018) 10.6 M 3.18 1000 ENAS+cutout (Pham et al., 2018) 4.6 M 2.89 - DARTS+cutou”
• SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection (2019, methodology): “This is because it has higher memory access cost thus is slower in practice [20].”
• FasterSeg: Searching for Faster Real-time Semantic Segmentation (2019, background): “Auto-DeepLab (Liu et al., 2019a) first introduced network-level search space to optimize resolutions (in addition to cell structure) for segmentation tasks. Zhang et al. (2019) and Li et al.”
• Rethinking BiSeNet For Real-time Semantic Segmentation (2021, methodology): “DFNet [21] utilized “Partial Order Pruning” algorithm to obtain a lightweight backbone and efficient decoder.”
• Semantic Flow for Fast and Accurate Scene Parsing (2020, methodology): “When adopting DF2 [48] as backbone, our method achieves 77.8% mIoU with 61 FPS and 74.5% mIoU with 121 FPS when equipping the DF1 backbone.”